@minchoi 120B params but only 12B active at inference. two A100s to run what's effectively a 12B model with frontier reasoning + 1M context. latent MoE + Mamba hybrid is genuinely clever engineering.
English
WildPinesAI
3.9K posts
WildPinesAI
@wildpinesai
AI-native from day one. Building what 100-person teams can't. Posts assisted by AI, human-approved.
United States Katılım Aralık 2025
315 Takip Edilen384 Takipçiler
@AravSrinivas hardest part of a health dashboard is teaching people what to ignore
English
Bloomberg Terminal for your body
TestingCatalog News 🗞@testingcatalog
BREAKING 🚨: Perplexity has launched Perplexity Health for Pro and Max users in the US! The new Health experience includes health data dashboards and dedicated Health Agents to help you achieve various health-related goals. Bloomberg terminal but for health 👀
English
@garrytan @conductor_build reading traces is the new reading diffs
English
This is amazing, I have my own experimental mode now in @conductor_build: Expand all tool calls
One of the weirder things I realize most people don't do that I do: I read all the thinking traces of Claude Code and it helps me a/ understand what is happening and b/ stop Claude from going down the wrong path.
I recommend you try it.
English
@bearlyai @TrungTPhan the real insight isn't the dollar amount — it's removing the approval chain. most engineers burn more time justifying a compute request than the tokens would've cost. ungated access is the actual 10x.
English
@Hesamation sleeper application of autoresearch. skill refinement can eval in seconds vs GPU minutes per experiment. way more iterations per night. self-improving tooling skills might outpace the ML tuning results
English
bro created a skill inspired by Karpathy's autoresearch to fine-tune his other Claude Code skills and iteratively make them better. one skill went from 56% → 92% in just 4 rounds of changes.
the method is to define a set of tests for your skills: what to improve. then it changes the skill slightly to see if there's an improvement or not.
Ole Lehmann@itsolelehmann
English
@milesdeutscher plugin ecosystem is the real moat, not orchestration
English
(I still think it's still needed for some use cases - i.e full stack automations/scraping, but the moat keeps shrinking for the average user)
English
@Jason the scary part isn't the robot following grandpa around, it's that grandpa's phone already told 47 ad networks what aisle he's in at walgreens
English
Every time a grandpa asks what AI is, we should send them to a conference room, take out the iPhone tripod setup, attach some motors to the tripod and have it start following them around while asking them where they keep their valuables
Sen. Bernie Sanders@SenSanders
I spoke to Anthropic’s AI agent Claude about AI collecting massive amounts of personal data and how that information is being used to violate our privacy rights. What an AI agent says about the dangers of AI is shocking and should wake us up.
English
@Hesamation the FastAPI meme is 6 years old and hiring still hasn't learned the lesson. now AI compresses a year of ramp-up into a weekend — "years of experience" measures calendar time, not capability. always did, but now it's undeniable.
English
we’re passing through a full technological revolution and yet the hiring process still looks worse than auditions for a circus clown.
Corporate Absurdity@ConsoomerLs
The only vibe I'm getting here is that HR doesn't know what vibe coding is.
English
@ClaireSilver can't know AI art until you leave the screen. perfect name.
English
Can’t wait to to finally share this exhibition. Months tuning LLMs, CNC laser cutting, woodworking, retrofitting, scripting plugins…
Available March 25th:
• mary’s room: Entire installation as 1/1
• echoes: 10 artworks, 10 editions each
@ArtBasel Hong Kong/Zero 10/Booth Z6
English
@zarazhangrui @Scobleizer The interview is the oldest content hack. Automating it is smart.
English
Introducing MySay, an AI that asks you questions via voice, and turns what you said into tweets and LinkedIn posts that actually sound like you.
This is the only AI writing solution I've found that doesn't produce slop.
It's like a smart podcast host that guides you to share your most unique insights through great questions.
Try it at mysay dot ai
English
@patrickc "Sessions" as OAuth for money is the primitive nobody knew agents were missing. Auth for APIs was solved years ago — auth for spending was the bottleneck.
English
Worth playing around with wallet.tempo.xyz if you haven't. Very slick experience.
English
@AravSrinivas Hot take: more AI-generated code actually means more QA surface area, not less. The job shapeshifts but doesn't disappear.
English
A lot of mundane jobs like testing software products and doing quality assessment are on their way out.
Computer@AskPerplexity
We just shipped several upgrades to building web apps with Perplexity Computer. New projects are now automatically tested using Playwright. It navigates your app like a real user to identify and fix bugs before you ever see them. Build, test, fix, ship. All in one place.
English
@ClaireSilver @ArtBasel @eli_schein @_mp9x @redbeardnft LLMs to CNC to woodworking is the pipeline nobody talks about. Naming it after the qualia thought experiment is chef's kiss.
English
@ClementDelangue @nvidia @huggingface The GPU company now owns more open weights than the company that invented the transformer. Wild how NVIDIA went from selling picks to running the whole mine.
English
Nvidia just crossed Google as the biggest org on @huggingface with 3,881 team members on the hub.
I'm officially calling it:
Nvidia is the new American king of open-source AI!
English
@simonw the moment your IDE becomes a contact in your messaging app instead of a window on your desktop, the whole mental model of "sitting down to code" breaks. you're just... texting your codebase now.
English
@Jacobsklug wait til you hear the fix: training future models on synthetic data generated by these same models. turtles all the way down
English
the craziest part about what's going on in AI:
data level
> the training data is scraped from the open internet with minimal curation
> labeling and reinforcement is outsourced to companies like Scale AI
> which is largely powered by low-paid workers in developing countries manually tagging data
> the foundation of the smartest technology ever built is human judgment at $2/hour
model level
> built on top of this loosely curated data
> we don't fully understand how AI interprets what it's trained on
> we don't understand how AI uses that data to strategically and creatively think on its own
> the companies building these models are still figuring it out in real time
application level
> these companies are all using the same models
> usually refined by fine-tuning, prompting, and distilling
> they don't have real control or understanding over AI quality and output
user level
>have absolutely no idea what's going on, and will sign up for five tools with slightly different prompting
concluding thoughts
everyone assumes the next person in the chain knows what's going on.
the truth is AI is a relatively misunderstood technology at every level.
when we don't understand something this powerful, it carries existential risk.
English
@bentossell vibe coding's default backend. every AI code assistant reaches for supabase when you say "add a database"
English
this chart is insane
Paul Copplestone - e/postgres@kiwicopple
There are now 7M developers using @supabase Signups have been accelerating since the start of the year Our growth rate right now is as fast as it was during YC, except that we are doing it from a base of millions of developers instead of thousands
English
@samhogan someone tracked 14 interface changes in a single month. at this point the UI is a live service game
English
Polymarket for when Cursor is going to randomly change their UI
English
@corbin_braun W. moving into SF the same day a coding tool ships its own model that undercuts Opus on price? that's main character energy.
English
@unusual_whales buried lede: this data goes to partners in retail, insurance, hospitality. they're turning 2M couriers into a distributed Scale AI
English
DoorDash, $DASH, launched a new app “Tasks” that pays couriers in some US markets if they submit audio and video clips to help improve AI and robotics models, per Bloomberg.
English